Simulation of Fire Smoke Diffusion and Personnel Evacuation in Large-Scale Complex Medical Buildings

Abstract

To address the significant problems of high fire risk and low evacuation efficiency in large and complex medical buildings, this study uses Ezhou Hospital as the empirical object to construct a multi-dimensional threat and risk assessment and fire evacuation dynamic coupling model and proposes a systematic optimization scheme to improve personnel evacuation safety. This study proposes an innovative full-chain analysis framework of “threat and risk assessment-dynamic coupling-multi-strategy optimization”. The specific methods employed include the following: (1) Using the probabilistic threat and risk assessment (PRA) method and the risk index (RII) method to identify the most unfavorable scenarios where the fire source is located in the outpatient hall (risk value C2 = 9.86). (2) Combining PyroSim and Pathfinder to construct a dynamic coupling model of fire smoke diffusion and personnel evacuation. Multiple groups, such as patients with mobility problems and rescue personnel, are added to address the limitations of traditional single-factor simulations. (3) Considering the failure of fire shutters, a two-stage optimization strategy is proposed for when the number of personnel is at its peak: the evacuation time is shortened by 23% by using internal intelligent guidance to shunt the congestion node crowd, and the addition of external fire ladders forms a multi-channel coordinated evacuation that further reduces the total evacuation time from 1780 s to 1266 s and improves the efficiency by 29%. The results show that the coupled multi-path coordination strategy and three-dimensional rescue facilities can significantly reduce the bottleneck associated with a single channel. This study provides a multi-dimensional dynamic evaluation framework and comprehensive optimization paradigm for the design of the evacuation of high-rise medical buildings and has important theoretical and technical reference values for improving the fire safety performance of public buildings and the intelligence of emergency management.

1. Introduction

The global aging population, chronic diseases, and intractable diseases are increasing, requiring centralized resources to provide efficient diagnosis and treatment; additionally, high-end equipment (such as proton therapy devices) require large, dedicated spaces [1]. In terms of public health emergencies, the COVID-19 pandemic highlighted the importance of large-scale infectious disease treatment capacities in large and complex hospitals [2]. There has been a significant increase in large-scale medical buildings around the world in recent years. China’s “14th Five-Year Plan” requires the expansion of tertiary hospitals, and more than 1000 tier 1 hospitals were built/renovated and expanded from 2021 to 2023 [3]. Europe and the United States integrate resources through the “medical city” model (such as the Texas Medical Center in the United States) [4]. Large complex medical buildings have complex functions (medical, scientific research, logistics), large equipment (precision instruments, oxygen supply systems), complex structural layering and zoning, high pressure on vertical transportation (especially tower stairwells), and multiple internal personnel (elderly, children, hospital bed patients, wheelchair patients, etc.). They also have a greater risk of fire, and evacuation of personnel is more difficult. The number of large hospital fires has remained high. For example, on 8 January 2024, the fire at the Diwaniyah Maternal and Child Health Hospital in Iraq resulted in the deaths of four children and injured more than 20 others [5]. On 12 December 2024, a fire at a private hospital in India killed at least seven people, including a boy. It took firefighters nearly two hours to put out the fire. All seven victims were found unconscious in the elevator and were sent to hospital for treatment, where they all died [6]. On 18 April 2023, a major fire accident occurred in Beijing Changfeng Hospital, causing 29 deaths and 42 injuries and a direct economic loss of CNY 38.3182 million [7]. On 24 September 2019, a fire broke out in a hospital in the city of Wade, the capital of the province of Wade in eastern Algeria, resulting in the deaths of eight babies due to burns and asphyxiation in the obstetrics and gynecology department [8]. Fire hazards in large and complex medical buildings have attracted extensive attention from local and international scholars, with the research focusing mainly on the following aspects:

(1)

The impact of various smoke exhaust measures on the flow of smoke

Selecting reasonable smoke exhaust measures can effectively reduce the threat posed by a fire. Duanmu et al. [9] showed that buildings can improve ventilation using different smoke exhaust methods, thereby increasing visibility during a fire. Zhong W et al. [10,11,12] analyzed the influence of mechanical smoke exhaust on the flow of fire smoke using numerical simulation and found that temperature has the greatest influence when the fire source is in the center of the building. Giachetti et al. [13] studied the interaction between buoyancy-induced upward flue gas flow and forced convection in mechanical ventilation using practical experiments and determined the optimal ventilation strategy in the event of a fire in a subway station. Lin et al. [14] found that due to mechanical smoke exhaust, air flow and heat exchange are more intense than natural smoke exhaust, and the attenuation coefficients for the heat release rate are 13% and 22%, respectively. They concluded that mechanical smoke exhaust is better than natural smoke exhaust. Pan [15], by studying effective smoke exhaust methods in subway stations and selecting different fire source locations to analyze the changes in temperature, visibility, and CO concentration, concluded that mechanical smoke exhaust plays a key role. In terms of water spray, Zhao et al. [16,17] investigated the inhibitory effect of water spray systems on the increase in heat release rate during combustion in a series of vertical fire suppression experiments in standard combustion chambers. Ma [18] applied the water spray system to fire smoke in a subway station, effectively reducing the temperature above the fire source and forming a low-temperature field within 400 °C above the fire source, reducing the high-temperature threat of the subway station ceiling. Bolina et al. [19] conducted a numerical analysis of buildings with variable ceiling heights under different fire scenarios, e.g., placing burners in the corners of the room and setting different heat output values. As the combustion process progressed, the fire released heat, which caused the surrounding air to heat up, causing the smoke from the combustion to spread upward. Xu et al. [20] used numerical simulation methods to systematically study the coupling relationship between flue gas diffusion dynamics and personnel evacuation efficiency in college dormitory fire scenarios, finding that comprehensive measures such as installing active sprinkler systems, implementing flow control of evacuees, and improving the structural design of smoke-proof partitions can effectively improve the necessary safe evacuation time of personnel. They also found that by optimizing the efficiency of evacuation routes, the time margin of ASET and RSET can be increased by 37.2%. Zhu et al. [21] jointly established a fire evacuation model of old buildings based on PyroSim using a large eddy simulation method and Pathfinder, simulated the influence of water spray and mechanical smoke exhaust on smoke exhaust indicators, and found that the effect of water spray was significantly better than that of mechanical smoke exhaust. Wang [22] showed that under the synergistic effect of mechanical smoke exhaust and water spray, the temperature drop rate was stronger than that of visibility and CO.

(2)

Emergency evacuation simulation

Li et al. [23,24] evaluated the effects of three escape staircase forms on personnel utilization and evacuation rate and found that evacuation efficiency was cross staircase > straight staircase > parallel double staircase. Zhang et al. [25] found, in the fire safety study of high-rise residential buildings, that closing the fire corridor can reduce the speed of smoke spreading directly upward through the corridor, but it will allow smoke to enter the stairwell; therefore, the rationality of the layout of the fire corridor should also be taken into account. Wang et al. [26] found, in an evacuation study of underground shopping mall fires, that the number of evacuees is positively correlated with evacuation time and the width of safe exits is negatively correlated with evacuation time, but the impact of evacuees on evacuation time is significantly higher than that of safe exit width. In the evacuation of personnel, the use of appropriate evacuation strategies has a significant impact on evacuation efficiency. Zhou et al. [27] classified various groups of people and then adopted effective evacuation methods for different groups, greatly improving evacuation efficiency. Choi et al. [28] used a simulation analysis of the social force model to verify that when a disaster occurs, leaders provide effective evacuation plans for personnel, indicating that providing effective information to those who need to evacuate can significantly improve the overall evacuation efficiency. Liu et al. [29] innovatively proposed a strategy similar to troop evacuation to simulate emergency evacuation in complex environments. Compared with unorganized evacuation, this strategy incorporates more environmental factors and significantly improves the adaptability to complex emergency environments. Chen et al. [30] proposed a multi-exit dynamic evacuation model that can significantly improve evacuation efficiency through intelligent allocation of evacuation routes and change rates, which is better than the no-guidance scheme. Zheng et al. [31], in his in-depth study of the problem of large-scale crowd evacuation, found that adding broadcast guidelines, signs, and planning evacuation routes for crowds can significantly improve evacuation efficiency, providing a powerful reference for solving the problem of large-scale crowd evacuation. In the planning research of the best evacuation path using algorithms, Liu et al. [32] developed an optimized artificial fish swarm algorithm that significantly enhanced the ability to solve multi-path planning problems by introducing a dynamic path evaluation mechanism and a group coordination strategy and realized the efficient exploration and global optimization of multi-objective routes in complex scenarios. Goel et al. [33] innovatively integrated the ant colony algorithm with the firefly algorithm through the cooperative optimization of the pheromone dynamic update mechanism and the light intensity guidance strategy, effectively addressing the local convergence dilemma in traditional path planning and significantly improving the global convergence accuracy of multi-objective path search in complex environments. Zhao et al. [34] developed an asynchronous dynamic programming mechanism by fusing the Dyna-Q framework with the tabular reinforcement learning algorithm, effectively solving the fast search problem of the optimal evacuation path in high-dimensional states in limited space.

(3)

Safety evacuation assessment

In the modern building fire protection design system, safety evacuation performance analysis and design, as the core component of performance-based fire protection design, occupies a pivotal position, especially for large hospitals with high population density and the complex personnel composition of public buildings. Effective safety evacuation performance analysis and design are of great significance in reducing the risk of casualties in fire accidents.

Pan et al. [35] conducted in-depth research on the key perspective of safe evacuation and innovatively proposed a series of optimization measures, including expanding the natural lighting area of the ladder, which provided a practical solution for improving the efficiency of safe evacuation in underground space. Li et al. [36] conducted detailed observation and analysis of the walking path and behavior of special groups in the Shanghai Rail Transit hub station in an experimental study of path exploration of special groups in the station and proposed measures such as widening the key channel and optimizing the marking system to make it more recognizable and guiding to ensure the evacuation safety of special personnel. Xie et al. [37] used a coupling analysis method combining agent behavior with computational fluid dynamics (CFD) models to explore the effects of fire location, fire heat release rate, crowd density, and escalator operation mode on personnel evacuation efficiency, finding that fire location and crowd density have the most significant impact on personnel evacuation. Liu et al. [38,39] used simulation methods to focus on measuring the temperature conditions, environmental visibility levels, and effective dose fractions of CO, CO₂, and O₂ during evacuation, providing a reference for safe personnel evacuation. Evacuation safety depends on two key time intervals: the available safe time and the necessary safe time [40,41]. The current evaluation system for building safety evacuation systems takes “the available safe evacuation time must exceed the necessary safe evacuation time” as the core criterion, while numerical simulation technology can not only reproduce the development of fire plumes by constructing a multi-physics field coupling model but also quantify the dynamic relationship between the available safe evacuation time and the necessary safe evacuation time in different scenarios. It has now become a core tool for optimizing building fire protection design and evacuation strategies [24,39,42,43,44,45].

In summary, previous studies have made significant progress in the field of fire safety in medical buildings, focusing on the verification of the effectiveness of smoke exhaust measures, the optimization of evacuation path planning algorithms, and the construction of security evaluation systems. However, existing studies are mostly limited to single-factor static analysis, and insufficient attention is paid to the real-time coupling mechanism of fire dynamic evolution and personnel evacuation behavior. There is a lack of refined modeling for the evacuation needs of multiple populations in complex medical buildings (such as patients with mobility problems and rescue workers), and there are few systematic solutions that combine multi-dimensional threat and risk assessment with multi-strategy collaborative optimization.

This study takes Ezhou Hospital as an empirical object and proposes an innovative full-chain analysis framework of “threat and risk assessment-dynamic coupling-multi-strategy optimization” using the probabilistic threat and risk assessment method (PRA) and risk index method (RII) to accurately identify the most unfavorable fire scenarios (personnel fail to extinguish the fire, but the water sprinkler system takes effect), and builds a fire evacuation dynamic coupling model based on PyroSim and Pathfinder, addressing the limitations of traditional single-factor simulation. In terms of evacuation, the evacuation situation where the fire compartment fails due to the failure of the fire shutter and personnel evacuation in the peak period is proposed. A two-stage optimization strategy is further proposed. Through the synergy of internal intelligent diversion and external three-dimensional rescue facilities, evacuation efficiency is significantly improved, providing a theoretical paradigm for multi-dimensional dynamic evaluation and comprehensive optimization for the study of fire safety in complex medical buildings.

2. Materials and Methods

2.1. Software Selection and Validation

The actual building fire evacuation drill has significant limitations. First, the drill is expensive and requires a lot of manpower and site resources. It is also difficult to reproduce real fire scenarios in complex medical buildings. Second, the number of experiments is limited and the environmental conditions (such as smoke diffusion path and personnel density distribution) are difficult to accurately control, resulting in insufficient repeatability and comparability of results. Third, the interaction between fire smoke diffusion and evacuation behavior cannot be dynamically quantified in actual scenarios, and it is difficult to comprehensively evaluate evacuation efficiency under multi-physics coupling. In order to address the above limitations, this study uses PyroSim (2022.3.1208) and Pathfinder (2024.2.1209) to construct a fire evacuation dynamic coupling model. PyroSim is based on the Fire Dynamics Simulator (FDS) framework and uses the large eddy simulation (LES) algorithm to solve the Navier–Stokes equation. It can accurately simulate the three-dimensional flow of fire smoke, the distribution of temperature fields, and the diffusion of toxic gases such as CO/CO₂. The meshing and boundary conditions are set in strict accordance with international standards (such as NFPA 92 [46]) to ensure the scientific appropriateness of fire dynamics simulations. Pathfinder can simulate individual behavior during a fire, taking into account various factors (e.g., channel width, walking speed, density), ensuring consistent results in the virtual environment and exploring the impact of different factors on evacuation efficiency, thereby avoiding biases in real-world drills.

PyroSim’s effectiveness demonstration: The fire research team at VTT’s Finnish Technical Research Centre used PyroSim to simulate a reconstruction of a wood stack fire test conducted in 2007 to verify the software’s effectiveness and practicality [47]. Using a cone burner, Wang [48] compared the simulation results of the large-size model and the small-size model of the training material with the experimental results. The results showed that the PyroSim software is effective and accurate at simulating the combustion heat release rate of high-speed trains. Su [49] compared the results of the scale model test and the numerical simulation calculation test and found that the two research methods gave similar results in terms of temperature above the fire source center, longitudinal temperature, and smoke spread morphology. In summary, PyroSim has significant effectiveness in numerical simulation and security evaluation of complex fire scenarios, verifying the reliability and practical value of the software in fire dynamics analysis and prevention and control strategy optimization.

Reliability demonstration of Pathfinder: Chen et al. [50] and Cao et al. [51] confirmed the reliability of Pathfinder in predicting evacuation speed, simulating individual behavior, and evaluating escape scenarios under smoke interference through a combination of experiments and modeling. Ren et al. [52] used channel evacuation test data and Pathfinder to reproduce the evacuation process and compared the simulation results with the test data. They showed that at different exit widths, especially at 1.3 m, 1.6 m, and 1.8 m, the Pathfinder simulation value had a small deviation from the test value, and the error control was good, verifying its reliability in evacuation scenarios. Cuesta et al. [53] further verified the reliability of Pathfinder in school evacuation scenarios. The simulation results were highly consistent with the actual evacuation data, and the average error of key parameters was controlled within 5%, confirming the accuracy of the model in the evacuation simulation of complex educational sites. In summary, Pathfinder has significant effectiveness in personnel evacuation behavior modeling and emergency scenario analysis and also shows high accuracy in evacuation simulation of complex building environments. The error of key parameters is controlled within 5%, fully verifying the engineering reliability and scientific credibility of the tool in diverse scenarios.

2.2. Model Building

Ezhou Hospital is located in Ezhou City, Hubei Province, and covers approximately 152,912 square meters. It is a tertiary hospital with a total of nine floors. The first floor is an emergency and outpatient hub, the second floor is an internal medicine outpatient and basic medical department, and the third floor is a surgical outpatient and specialty outpatient department. The first to third floors are located in the podium. The surgical and maternity and children’s buildings are located on the left and right sides of the tower on the fourth to ninth floors, including inpatient areas, surgical areas, and maternity and child buildings. The specific model is shown in Figure 1.

2.3. Fire Parameter Setting

2.3.1. Heat Release Rate Setting

This study applies to a public building. According to the “Technical Standard for Smoke Management Systems in Buildings” GB51251-2017 [54], the heat release rate is 2.5 MV under the consideration of water spray.

The relationship between the heat release rate and time conforms to the following formula (the t² model is used to analyze the correlation between the two in non-laboratory simulations).

Q = α t²

(1)

where Q is the heat release rate of the fire source; α is the fire growth coefficient, in kW/s²; and t is the time required for the fire source to reach the heat release rate. According to Equation (1), the time required to reach the maximum heat release rate is 731 s.

2.3.2. Selection of Combustion Materials

In the choice of combustion materials, polyurethane is used as the main component. Due to its significant effect in improving patient comfort and preventing bedsores, polyurethane is widely used in hospital mattresses, medical seats, pads, etc.; however, it has also become a safety hazard that cannot be ignored due to its flammability. In fire simulations, polyurethane, as a combustion material, can simulate a fire scene with fierce fire, thick smoke, and a lot of toxic gases, which is close to the actual situation.

2.3.3. Grid and Observation Point Settings

At the key nodes of the evacuation path (inside each exit door frame), a human feature height monitoring plane 1.6 m away from the floor slab is set to dynamically monitor CO concentration, temperature, and visibility. After the grid independence test, the grid size of the model is set to 1 m × 1 m × 1 m, and the number of grids is 1,254,619. The location of the detection points is shown in Figure 2.

2.3.4. Fire Available Safety Time Determination

According to the American Society of Fire Engineers’ reference temperature tolerance times, the human body can tolerate temperatures of 60–100 °C for only 12 to 30 min. According to the “Building Smoke Prevention and Exhaust System Technical Standard” GB51251-2017, “In terms of safe evacuation, the visibility in the fire field should not be less than 10 m for larger spaces, and in smaller spaces, the visibility should not be less than 5 m”. The higher the familiarity of personnel with the environment, the lower the visibility requirements. Therefore, the visibility at the characteristic human height of 1.6 m is 5 m, making this the hazard judgment standard. The SFPE Fire Engineering Manual describes human contact time in a CO environment: humans can tolerate a CO concentration of 1000 ppm for 30 min. Considering the nature of the fire source in this study and the time required for the fire to spread in the building, the CO exposure threshold is taken to be 500 ppm. The CO concentration hazard determination index is greater than or equal to 500 ppm. These three indicators are used to determine the available safe time [24] as follows:

$$\mathrm{A}\mathrm{S}\mathrm{E}\mathrm{T}=\mathrm{m}\mathrm{i}\mathrm{n}\left(t_{CO},t_{Visibility},t_{Temperature}\right)$$

(2)

where $t_{CO},t_{Visibility},t_{Temperature}$ are the times required to reach the separate thresholds.

2.4. Personnel Evacuation Parameter Setting

2.4.1. Personnel

According to the CAD drawings and on-site research, the maximum number of people in this building is 1200 on the first floor, 600 on the second and third floors, and 1746 and 1689 on the left side of the fourth and ninth floors, respectively, giving a total of 5835 people. In terms of personnel distribution, combined with the characteristics of the hospital and on-site research, the first and second floors of the outpatient clinic have the highest flow of people, focusing on key first aid and diversion functions. Among them, mobility-impaired people are more likely to cause personnel blockage due to physical mobility difficulties, which also brings greater fire risk; therefore, on these two floors, there are a total of 1800 people, and the ratio of mobility-impaired people to mobility-unimpaired people is 3:2 [55]. Based on the above assumptions, the number of mobility-impaired people is 1080, including 300 children, 300 wheelchair-bound patients, 360 elderly people, and 120 bedridden patients. Due to the presence of the medical technology department on the third floor of the hospital, which also serves as a connecting floor with the tower, 80 beds are considered; thus, there are 80 bedridden patients and 136 hospital staff. Due to the fact that most of the hospital staff are nurses with frequent night shifts, women have an advantage when it comes to dealing with this meticulous and stressful work. According to on-site research, the male/female ratio of nursing staff is about 1:3; that is, 34 men to 102 women, and the remaining 384 are adult patients and their families. There are 1689 people in the surgery wing on floors 4–9, and the proportion of medical staff is slightly higher than in the podium (22%; 384 people), of which the ratio of men to women is 1:3 (96 and 288, respectively). Of the remaining 80% of patients, 115 are bedridden, 115 are wheelchair users, and 1075 are other patients and their families. There are 1746 people in the women’s and children’s building on the right side of floors 4–9. Considering the many women and young children and the need for more care, the proportion of medical staff is slightly higher than in the surgery building: 33% are medical staff (566 men and 432 women), and 70% are patients, including 115 bedridden, 115 wheelchair users, and 940 other patients and their families. The distribution of all patients and their families shows that the ratio of males to females is 3:7, as shown in Table 1 below, where the staircase distribution is Figure 3. The bed wheelchair push diagram is shown in Figure 4.

Table 1. Number of personnel types on each floor.

Floor	Adult Male	Adult Female	Children	In a Wheelchair	Bedridden	Old Men	Total
1	200	280	200	200	80	240	1200
2	100	140	100	100	40	120	600
3	106	174	60	60	80	120	600
4–9 left	403	595	115	115	115	346	1689
4–9 right	412	700	102	115	115	302	1746

Table 1 shows the specific distribution of personnel at all levels. Note: Adult males include male adult patients and male healthcare workers (rescuers), and adult females include female adult patients and female medical workers (rescuers). Of these, 17% of the nursing staff are on the podium, while 22% of the nursing staff are in the surgical section (left side of floors 4–9)—due to the fact that the slightly larger proportion of people with limited mobility and the women and children in the women and children’s building (right side of floors 4–9) need more care—thus, 33% of the people are nursing staff. The ratio of men and women in the nursing staff is about 1:3 [55], as shown in

Table 2. Composition of adults in hospital.

Floor	Adult Males		Adult Females
	Rescue Workers	Normal People	Rescue Workers	Normal People
1	52	148	156	124
2	26	74	78	62
3	34	72	102	72
4–9 Left	96	307	288	307
4–9 right	144	268	432	268

.

Table 1. Number of personnel types on each floor.

Floor	Adult Male	Adult Female	Children	In a Wheelchair	Bedridden	Old Men	Total
1	200	280	200	200	80	240	1200
2	100	140	100	100	40	120	600
3	106	174	60	60	80	120	600
4–9 left	403	595	115	115	115	346	1689
4–9 right	412	700	102	115	115	302	1746

Table 1 shows the specific distribution of personnel at all levels. Note: Adult males include male adult patients and male healthcare workers (rescuers), and adult females include female adult patients and female medical workers (rescuers). Of these, 17% of the nursing staff are on the podium, while 22% of the nursing staff are in the surgical section (left side of floors 4–9)—due to the fact that the slightly larger proportion of people with limited mobility and the women and children in the women and children’s building (right side of floors 4–9) need more care—thus, 33% of the people are nursing staff. The ratio of men and women in the nursing staff is about 1:3 [55], as shown in Table 2.

2.4.2. Personnel Attributes

In terms of the settings of personnel attributes, according to the “SFPE Handbook of Fire Protection Engineering” (SFPE Manual) [56] and empirical research, the speed of adult males is 1.2 m/s–1.6 m/s, and there is no physiological difference between adult females and adult males. The speed of children is 0.9 m/s–1.3 m/s, that of the elderly is 0.7–0 m/s, that of single wheelchair users is about 0.9 m/s, and that of two people lifting beds is about 0.8 m/s. In terms of shoulder width, adult males are 45–50 cm wide, women are slightly narrower, about 45 cm wide, and the elderly may be narrower, about 42.5 cm wide, due to factors such as hunchback (

Table 3. Personnel attributes.

Personnel Classification	Speed (m/s)	Shoulder Width (cm)	Height (m)
Adult male	1.45	48	1.7
Adult female	1.4	45	1.6
Child	1.2	34.2	1.35
Person in wheelchair	0.9 (Auxiliary push speed)	50 (Wheelchair width)	1.7 (Adult male)
Immobile person	0.8 (Assisted lifting speed)	100 (Bed width)	1.6 (Adult female)
Old man	0.8	42.5	1.63

).

2.5. Selection of Fire Scenarios

2.5.1. Fire Probability

Based on the basic principles of probability analysis composed of mathematical statistics theory, system safety theory, and probability theory, this study aims to determine the probability of the occurrence of each state during a fire in Ezhou Hospital to address the uncertain factors, such as the response efficiency of fire equipment at fire scenes and the opening and closing of fire compartments. It also constructs a threat and risk assessment framework to realize the quantitative assessment of fire risk in Ezhou Hospital.

The effectiveness of fire protection systems is based on the British standard BSDD240 “Fire safety engineering in buildings” [57], and the potential failure risks of different types of fire protection facilities are quantified using statistical methods. The specific data are shown in

Table 4. Statistics of effectiveness of each type of fire protection equipment.

Type of Fire Protection Equipment	Invalid Probability
Water spray system	0.10
Mechanical smoke exhaust system	0.15
Fire partition	0.3

. The probability of ineffectiveness reflects the possibility that the relevant facilities cannot perform their design functions normally in a fire scenario. The numerical difference is due to the technical characteristics of the facility, the maintenance complexity, and the environmental dependence of the facility. The probability of successful fire suppression by emergency personnel is given by a p-value between 0 and 1; the higher the p-value, the greater the probability of occurrence. Considering that there are many staff in the hospital hall, the emergency fire extinguishing capacity is relatively higher; thus, the p-value is set to 0.9.

Based on analyses of previous hospital fire accidents, personnel are often concentrated in the hall, and the hall on the first floor is the main entrance and exit, with a large number of personnel and a more complex composition; therefore, the risk of fire is greatest in the hall, and the consequences are also more serious. As shown in

Table 5. Probability and risk statistics for each scenario.

Fire Location	Personnel Put out the Fire in a Timely Manner	Water Spray System	Mechanical Smoke Exhaust System	Fire Partition	Event Probability	Risk Statistics
Lobby	Yes (p11a = 0.9)	\	\	\	S1 = 0.9	C1 = 0
	No (p11b = 0.1)	Yes (p12a = 0.85)			S2 = 0.085	C2 = 9.86
		No (p12b = 0.15)	Yes (p13a = 0.9)	\	S3 = 0.0125	C3 = 3.64
			No (p13b = 0.1)	Yes (p14a = 0.7)	S4 = 0.001	C4 = 1.75
				No (p14b = 0.3)	S5 = 0.0004	C5 = 0.817

, in the fire table analysis for Ezhou Hospital, five different fire scenarios were constructed based on different working conditions. Scenario 1: The personnel successfully extinguished the fire in time; Scenario 2: the personnel did not extinguish the fire in time, but the water sprinkler system took effect and extinguished the fire successfully; Scenario 3: the personnel did not extinguish the fire in time, and the water sprinkler system also failed; Scenario 4: the personnel did not extinguish the fire in time and the mechanical smoke exhaust and water sprinkler systems failed, but the fire protection partition took effect and controlled the fire source; Scenario 5: all measures failed. By multiplying the probability of occurrence of each pre-influencing factor, the specific probability of occurrence of each fire scenario can be calculated.

2.5.2. Consequence Analysis

According to the “Fire Safety Engineering” GB/T 31592-2015 [58], the number of people threatened is used to estimate the consequences of each scenario. According to CAD drawings and on-site studies, the estimated total number of people is about 5835. Through previous research on a museum hall [59], when the fire source is extinguished by emergency personnel in time, the risk is negligible. When personnel fail to extinguish the fire in time, the water spray coefficient takes effect, and about 1% of people are threatened. When the water spray system fails, about 3% of people are threatened when the mechanical smoke exhaust system takes effect. When the above measures fail, the windproof zoning takes effect, and about 30% of people are threatened. Considering that the hospital hall contains more people with limited mobility, more serious consequences are likely due to congestion; therefore, the following assumptions are made. When the fire point is located in the hall, combined with the scene situation, a total of five scenarios are set. Scenario S1: The fire is quickly extinguished by emergency personnel, indicating that the scenario does not result in serious consequences; thus, consequence B1 = 0. Scenario S2: Due to the timely operation of the water spray system and the control of the heat release rate, 2% of people are threatened; thus, B2 = 5835 × 0.02 = 116. Scenario S3: Due to the effective absorption of flue gas by the mechanical exhaust system, 5% of the people are threatened; thus, B3 = 5835 × 0.05 = 291. In Scenario S4, the fire compartment has a controlling effect on the spread of flue gas, limiting its spread. Assuming that 30% of people are threatened, then B4 = 5835 × 0.3 = 1750. In Scenario S5, all fire protection means are ineffective. Assuming that 35% of the hall is threatened, B5 = 5835 × 0.35 = 2042.

The product of the probability of a scene occurring and the consequences of a scene is the relative risk [59]. As shown in Equation (3), the risk statistics of the probability of a scene occurring and the number of threatened people in the scene are recorded as C1–C5. According to the risk statistics, we select Scenario C2 as the fire research scenario. Table 5 shows the probability and risk statistics of each scenario.

C = S × B

(3)

where C is the risk of the scenario, S is the probability of the scenario occurring, and B is the number of casualties in the scenario.2.6. Evacuation Scenario Selection

As shown in Figure 5, there are a total of nine fire compartments on the first floor of this building. Different fire compartments are separated by walls and fire shutters. In the event of a fire, personnel in each fire compartment use the evacuation channels of the corresponding fire compartments to evacuate outdoors.

The internal evacuation diversion strategy is to reduce the waiting time by guiding the crowd to different exits or routes to avoid overloading a single exit by combining the principle of channel capacity balance in evacuation dynamics as follows:

$$J_{max}={\rho }_c·v_0·W$$

(4)

where $J_{max}$ is maximum traffic; ${\rho }_c$ is the critical density (usually 1.5 minus 2 divided by m²); W is the stair width; and $v_0$ is the average speed of personnel. When $\rho >{\rho }_c$, the stairs enter a crowded state, triggering a personnel diversion strategy. The ideal situation is Ci (diverted channel) = Cj (diverted channel), so that the flow of people in each evacuation channel is equivalent, achieving maximum evacuation efficiency.

Internal evacuation diversion can avoid the hedging of up and down crowds (such as the overlap of rescue workers and evacuees) and improve the traffic efficiency of stairwells and corridors. When the internal diversion strategy reaches the ideal situation and still cannot meet the evacuation safety, the ladder provides an external vertical rescue path, which can once again divert the congested crowds to relieve congestion, especially in high-rise buildings (such as hospital towers and office buildings).

Considering fires that have occurred, the Harbin warehouse fire in China [60], the Russian shopping mall fire [61], and the Daegu subway arson case in South Korea [62] all showed that some roller shutters do not work normally due to improper maintenance or faulty automatic control systems. Failure to respond to fire signals in a timely manner caused the failure of fire compartments and resulted in serious consequences, such as a large number of casualties.

Based on the relevant literature [50,63,64] and the on-site situation, the number of people in the trough period is about 75% of those in the peak period. This scenario, where the number of people is 75% of the maximum number (peak period), is analyzed at the same time.

We set up two optimization strategies for normal evacuation in cases of effective fireproof partition and failure of fireproof partition, as shown in

Table 6. Specific settings for five working conditions.

Working Condition	Fire Partition	Evacuation Strategy
1	take effect	Evacuate the evacuation channel corresponding to the fire compartment to the outside
2		75% personnel
3	failure	Normal evacuation
4		Internal evacuation diversion
5		Building a safety ladder on the roof of the podium on the basis of internal diversion

.

3. Results

3.1. Fire Safety Analysis

The core of fire safety analysis lies in quantifying the comprehensive impact of fire dynamics on building spaces and occupant safety. To systematically assess the risk characteristics of a fire scenario at Ezhou Hospital, this study selects the outpatient hall as the fire source benchmark and employs the PyroSim simulation platform for three-dimensional dynamic modeling of smoke dispersion. Critical monitoring point data are extracted to determine environmental hazard thresholds. By analyzing smoke flow paths, toxic gas concentration distributions, and visibility attenuation patterns, high-risk areas and evacuation bottlenecks within the building can be precisely identified. This study first focuses on the spatiotemporal evolution mechanisms of smoke dispersion to reveal the direct impact of fire environments on occupant evacuation, thereby providing a scientific basis for subsequent evacuation time margin analysis.

3.1.1. Flue Gas Diffusion Analysis

We performed a flue gas simulation based on the riskiest scenario, Scenario 2 (water spraying). Figure 6 shows the flue gas diffusion at 1500 s.

According to Figure 6, at 1500 s, the horizontal corridors and turns are completely filled with smoke, and the thickness of the smoke layer is significantly increased; at the same time, the lower left of the first floor is about to be filled with smoke, and the smoke is about to diffuse into the lower right of the first floor. Half of the lower part of the first floor is also covered by smoke. The concentration of smoke in the corridor leading to the staircase is high. From a vertical perspective, the third floor of the podium has been completely filled with smoke, and the stairwell leading to the lower floor has also begun to be filled with smoke. The surgical building on the left side of the tower is extremely affected by smoke. The smoke has begun to spread down from the top floor. The stairwell on the right side of the tower is slightly affected.

In the fire scenario in this study, polyurethane is the main combustion material and its high carbon chain structure is prone to generating large numbers of incomplete combustion products (such as CO and soot particles) during the pyrolysis process, resulting in a significant increase in flue gas density, further reducing environmental visibility and exacerbating evacuation risks. Overall, the flue gas diffusion shows obvious temporal and spatial laws and path dependence. In the initial stage, the flue gas rapidly accumulates near the fire source and spreads to the roof. Under the action of buoyancy, a vertical updraft is formed, and it spreads to the periphery through the horizontal corridor, and local flue gas infiltration begins to appear in the staircase. In the medium stage, the flue gas forms a two-way diffusion path through the horizontal corridor and the vertical staircase, and the podium is gradually covered by the flue gas. The “chimney effect” accelerates the migration to the upper part. In the later stage of the fire, the smoke fills the horizontal corridor on the first floor and spreads to the high-rise surgical building through the stairwell on the left side of the tower, while the women’s and children’s building on the right is less affected by the smoke due to the spatial separation effect. The overall diffusion path shows characteristics of “horizontal spread dominates, vertical accelerated penetration”, and the functional zoning and structural layout of the building significantly affect the distribution of the smoke.

3.1.2. Determination of Available Safety Time

Monitoring point 1 closest to the fire source is selected as the reference standard for the available safety time around the fire source location. The CO, temperature, and visibility curves of monitoring point 1 are shown in Figure 7.

Monitoring point 1 is located at the height of human eye detection, at a vertical distance of 1.6 m from the gate closest to the fire source. The safety of personnel evacuation can be generally judged by the changes in various indicators at monitoring point 1. According to Figure 7, the temperature of monitoring point 1 rises rapidly with time, briefly reaches 60 °C at 400 s, and then remains between 50 °C and 60 °C. The maximum CO reaches 70 ppm, which is far less than the safety threshold of 500 ppm. The visibility decreases rapidly to about 5 m at 0–400 s. With the influence of smoke movement, the visibility varies in the range of 5 m–30 m at 400 s–1400 s. After 1400 s, the visibility drops to within 5 m. Based on the above conditions, the available evacuation time for Ezhou Hospital is about 1400 s.

3.2. Total Evacuation Time for Different Evacuation Strategies

Based on the simulation results of the fire dynamics coupling model, this section systematically evaluates the effectiveness of different evacuation strategies for the most adverse fire scenario in the outpatient hall (risk value C2 = 9.86). The study focuses on five typical conditions: conventional evacuation with active fire compartments (Strategy 1), off-peak personnel diversion (Strategy 2), unstructured evacuation in the event of fire compartment failure (Strategy 3), and a composite strategy integrating internal path optimization with external aerial ladder assistance (Strategies 4–5). By comparing total evacuation time with available safe time, the analysis highlights evacuation bottlenecks under extreme conditions, such as fire compartment failure and peak-hour congestion. Furthermore, it verifies the effectiveness of multi-path coordination in enhancing passage capacity at critical nodes. A quantitative analysis is conducted from the perspectives of evacuation time, personnel distribution heatmaps, and exit flow rates.

3.2.1. Conventional Evacuation with Active Fire Compartments (Strategy 1)

Under normal circumstances, the first floor fire compartment takes effect, and the personnel in each fire compartment are evacuated outdoors via the evacuation passage corresponding to their fire compartment. The personnel evacuation simulation results are shown in Figure 8.

Using Pathfinder to analyze this scheme shows that the evacuation time for personnel is 1710 s (Figure 8), which is significantly higher than the available safe time of 1400 s; thus, evacuation optimization is required. Since people on the second and higher floors can freely choose their evacuation path during this evacuation process, congestion still occurs on the stairs, especially in the later stage of evacuation. The slope of the curve is too low, indicating that there is a lot of congestion in the stairwell.

3.2.2. Off-Peak Personnel Diversion (Strategy 2)

Under normal circumstances, the fire compartment is effective, but the hospital operation is in a trough period, and the personnel are only at 75% of the peak period. The simulation analysis results are shown in Figure 9.

Pathfinder is used for analysis based on this scheme. According to Figure 9, the evacuation time is about 1400 s, which is equivalent to the available safe time of 1400 s for fires. Evacuation still has a certain risk; therefore, evacuation optimization is required. Due to the number of people being fewer than at the peak period in this strategy, the slope of the second half of the curve is lower than the working condition 1, indicating that the congestion in the later stage is slightly better than in working condition 1; however, the degree of congestion is still high at 1350 s–1400 s in the later stage of evacuation.

3.2.3. Unstructured Evacuation in the Event of Fire Compartment Failure (Strategy 3)

For this strategy, we based the analysis on the above assumptions of personnel composition and attributes under the failure of the fire partition and the extreme condition that the number of personnel is at the maximum, the personnel are unorganized, all conventional (non-fire-fighting) doors are open for use (by default), and people use the nearest or most familiar exits. There is no priority on the staircase, and no priority evacuation path is specified. The evacuation simulation results are shown in Figure 10.

According to Figure 10, all 5835 people in the hospital were evacuated over the 1780 s. Efficiency was highest in the 262 s at the beginning of the evacuation due to the small number of people in the staircase and the low congestion. Most of the people who evacuated were normal people. This is because, at this time, the hospital bed personnel and wheelchair personnel were waiting in place for rescuers, and the speed was slow. The evacuation efficiency rapidly reduced in the 262–1200 s period in the middle of the evacuation due to the concentration of personnel on the staircase, the presence of rescuers and people with limited mobility, and hospital beds and wheelchairs occupying part of the space on the stairs affecting the evacuation of the crowd. At the same time, the stairs were occupied by the rescue personnel (firefighters, healthcare workers) and the evacuated people going down, resulting in hedging congestion. Due to a lack of broadcast guidance and people blindly following others to the same paths, a “herd effect” formed, leading to crowding on some stairs and leaving other stairs vacant for a long time, resulting in low utilization efficiency. This period is also the most prone to serious events, such as stampedes. The remaining 300 people were evacuated in the 1200–1780 s period, and the curve in the time period map is close to the horizontal line, indicating that the evacuation efficiency is extremely low and there may be a severe blockage in the stairwell.

Figure 11 shows that the evacuation path distances of the main population of 5835 evacuees from the hospital were concentrated in the range of 0–150 m, and the evacuation was completed within 750 s. Further analysis found that the evacuation time of the remaining few individuals was significantly longer (>1200 s), but the increase in the path distance was minimal, indicating that the time lag was mainly attributed to the local congestion effect of key nodes rather than the path length itself.

Figure 12 shows the statistics of the flow of people through key exits. Data analysis reveals significant differences in pedestrian flow rates at each exit. An analysis of several entrances and exits shows the following: Figure 12b shows that the main entrance is affected by the fire. After the crowd in the hall is quickly evacuated in the early stage of the fire, no one evacuates through the entrance. Figure 12a,f indicates that Stair S-1 and Stair S-18 are located near the main entrance, and some people pass through the podium in the early stage, but no one passes through later. Figure 12c shows that Stair S-5 on one side of the tower has an obvious effect on the evacuation of the crowd at the podium. People continuously pass through this exit within 1000 s. Figure 12d,e indicates that Stairs S-9 and S-10 in the tower area exhibit a significant herd effect: a large number of people gather at these stairwells in the initial stage; as evacuation progresses, a large number of people still gather at these stairwells, resulting in long evacuation times and low evacuation efficiency.

The above analyses show that in the scenario of unorganized evacuation, the total evacuation time (TET) calculated through computer simulation and the human behavior model is approximately 1780 s, which significantly exceeds the critical value of 1400 s for the available safe egress time (ASET) in a fire scenario. According to the standard regulations on the safety margin of evacuation time [65], there are major potential safety hazards when TET/ASET > 1.

The existing evacuation plan does not meet the requirements of the building fire protection code, and it is necessary to improve evacuation efficiency, e.g., by optimizing the evacuation routes and installing external evacuation ladders.

3.2.4. Internal Evacuation Path Optimization (Strategy 4)

To solve the above problems, it is necessary to specify a reasonable and effective evacuation strategy. Considering that one of the reasons for the low evacuation efficiency is the conformity of personnel in the selection of evacuation stairs, the crowd in the congested staircase section is diverted to other stairs through internal optimization to reduce evacuation pressure on a single staircase. The results of the simulation of this strategy are shown in Figure 13.

According to Figure 13, evacuation time after personnel-guided evacuation was reduced from 1780 s to 1370 s, and the evacuation efficiency increased by 23%. The smoothness in the second half of the curve was significantly weaker than that of unorganized evacuation, indicating that the evacuation strategy effectively alleviated the congestion problem in the later stage.

S9 and S8 are the most congested and unobstructed stairs on the surgical building side, respectively. The total evacuation time can be reduced by moving people from the more congested stairs to the more unobstructed stairs. Figure 14 shows the flow rate of personnel through the S8 and S9 stair exits. After guiding the evacuees, the evacuation speed through both staircases increases. Stairs S-8 and S-9 have similar personnel evacuation times. Some people who originally went to the S-9 stairs were evacuated via the S-8 stairs. Evacuation monitoring data show that the efficiency of the building evacuation system is significantly improved by implementing the directional guidance strategy. Dynamic analysis of the flow of people at the S-8 and S-9 stairways in the key observation areas shows that personnel diversion makes S-8 bear the original evacuation flow of S-9, the evacuation times of the two stairs tend to be balanced, and the overall evacuation efficiency of the system improves, which verifies the effectiveness of the load balancing strategy.

Therefore, dispersed guidance of the crowd and the average use time of each staircase are effective in improving evacuation efficiency. The total evacuation time is 1370 s; although this is less than the available safety time of 1400 s (TET/AEST < 1), considering that the personnel have information on the fire during the evacuation process and the time from learning about the fire to the reaction time of personnel is known, the safety during evacuation is still not high and needs to be optimized further.

3.2.5. External Fire Ladder Optimization (Strategy 5)

Since the early guidance measures have ideally balanced the distribution of personnel at each staircase, the effect of continuing to strengthen guidance on improving evacuation efficiency is limited. Therefore, we propose a plan to add escape ladders on the facade of the building to open up a new evacuation path. According to the structural characteristics of the building, two ladders are selected for deployment on the roof of the podium (see Figure 15 for the specific location). This design not only effectively utilizes the characteristics of the building space but also forms a multi-point evacuation pattern that significantly improves the chances of personnel escape.

In the design of evacuation behavior mode, personnel move to the roof platform through the podium entrance and exit and then initiate the escape procedure using the ladder. Due to the response delay characteristics of the rescue equipment (the deployment and debugging of the ladder takes about 150 s), the escape channel is set to activate at t = 150 s in the numerical simulation model.

Simulation analysis was performed based on the above conditions. The results are shown in Figure 16.

According to Figure 16, the quantitative evaluation results show that the ladder intervention reduces the overall evacuation time from 1782 s to 1266 s and improves the efficiency by 29%. This improvement is due to the synergistic effect of multiple channels, indicating that the introduction of auxiliary escape facilities can effectively address the bottleneck caused by a single channel.

As shown in Figure 17, some personnel can be evacuated using the self-ladder after it is erected. Since it is necessary to walk a certain distance from the entrance of the podium to the ladder, and because the podium roof is wider, personnel congestion can be alleviated. The evacuation of personnel via the ladder can be completed at a faster speed. Since the crowd at S-9 can be further diverted, congestion on the stairs can be reduced, making the speed of personnel evacuation relatively stable. This shows that after the internal evacuation path is optimized, the external safety ladder is used to form a multi-point evacuation pattern, which is further optimized. The evacuation time is 1266 s, which is significantly less than the safe time available in a fire (1400 s); thus, the use of safety ladders is effective.

4. Discussion

This study constructs a fire evacuation dynamic coupling model to reveal the interaction mechanisms between smoke diffusion and occupant evacuation in large medical building fire scenarios and proposes a multi-strategy collaborative optimization approach. This section discusses the findings from two perspectives: comparison with existing studies and study limitations.

4.1. Comparison and Innovation with Existing Research

This study overcomes the limitations of traditional single-factor analysis by implementing a dynamic coupling model that enables real-time interactive simulation of fire evolution and evacuation behavior.

For instance, in the smoke diffusion analysis, it was found that while the CO concentration produced by polyurethane combustion did not exceed the threshold (maximum 70 ppm), visibility dropped to a hazardous level (<5 m) after 1400 s. This finding aligns with Xu et al.’s [20] study on university dormitory fires, which concluded that visibility reduction is the dominant factor in evacuation risk. However, this study further quantifies the accelerated visibility degradation caused by the combustion of special materials in medical buildings, showing a 15–20% faster decline compared to ordinary buildings.

Furthermore, the refined modeling of mobility-impaired patients (such as wheelchair users and bedridden individuals) addresses the limitation of Zhou et al.’s [27] study, which categorized populations solely based on age. The findings reveal that the turning radius requirement of such individuals (wheelchair width: 50 cm) significantly hinders passage efficiency at stairway entrances, providing valuable data support for evacuation design tailored to special populations.

4.2. Research Limitations and Future Directions

Despite the achievements of this study, the following limitations remain:

(1)

Practical Constraints of External Rescue: The deployment of fire truck ladders requires a response time of 150 s. However, in real fire scenarios, equipment adjustments may be affected by environmental factors such as high temperatures and smoke obstruction, leading to increased response delays. It is recommended to integrate drone inspections and intelligent scheduling systems to optimize the ladder deployment process.

(2)

Limitations in Data Validation: The evacuation parameters in this study were set based on historical literature and localized investigations, without full-scale real-person evacuation experiments. Future research could leverage Mixed Reality (MR) technology to construct virtual fire scenarios, enabling the collection of real human behavior data to enhance model accuracy.

5. Conclusions

To address the problems of high fire risk and low evacuation efficiency in large medical buildings, this study proposes an innovative full-chain solution of “threat and risk assessment-dynamic coupling-multi-strategy optimization”. The probabilistic threat and risk assessment method (PRA) and the risk index method (RII) precisely identify the outpatient hall as the most unfavorable fire source scenario, and a fire evacuation dynamic coupling model is constructed based on PyroSim and Pathfinder to realize real-time interactive simulation of smoke diffusion and personnel behavior.

In terms of evacuation, under normal circumstances, when the fire compartment takes effect, the time required for all personnel to evacuate is 1710 s. When only 75% of the personnel are in the trough period, the evacuation time is 1400 s, which is a more dangerous situation. When the fire compartment fails and personnel are in the extreme situation of the peak period, the evacuation time is 1780 s, which is more dangerous. The total evacuation time is reduced from 1780 s to 1266 s and the efficiency is increased by 29% through two-stage optimization (internal path intelligent guidance + external ladder coordination). This verifies the key role of the multi-channel coordination mechanism in addressing the single channel bottleneck. Figure 18 shows a comparison of evacuation times under five conditions.

(1)

This study combines probabilistic threat and risk assessment (PRA) with the risk index method (RII) to screen out the most unfavorable fire scenarios (risk value C2 = 9.86), where the fire source is located in the outpatient hall, and realizes the joint simulation of fire smoke diffusion and personnel evacuation through the dynamic coupling model of PyroSim and Pathfinder. The model solves the limitations of traditional single-factor analysis, accurately quantifies fire risk and evacuation efficiency, and provides a multi-dimensional theoretical framework and technical support for complex medical building security evaluation.

(2)

An evacuation path optimization scheme based on intelligent guidance is proposed, and a load balance shunt strategy is implemented for key congested nodes (S-9, S-10), which shortens the total evacuation time from 1780 s to 1370 s and improves efficiency by 23%. Through verification of the dynamic density heat map, the strategy effectively alleviates the “herd effect”, reveals the synergistic improvement of the spatial buffering effect and node shunt effect on evacuation stability, and provides an operable optimization paradigm for hospital evacuation design.

(3)

The addition of fire ladders to form a multi-channel coordinated evacuation network combined with the deployment of ladders with a response delay of 150 s further reduces the total evacuation time to 1266 s and increases the efficiency by 29%. The scheme significantly reduces the evacuation pressure in high-rise areas due to the synergy of vertical rescue paths and horizontal evacuation, verifying the key value of external three-dimensional rescue facilities in addressing the bottleneck of traditional single channels.

(4)

The dynamic mechanism of spatial buffer effect, path extension effect, and node shunt effect in evacuation optimization are revealed for the first time through dynamic coupling simulation, and a comprehensive optimization framework of “internal shunt + external aid coordination” is proposed. This mechanism not only provides a quantitative basis for the design of the evacuation of complex medical buildings but also lays a theoretical foundation for the integration and application of intelligent sensing technology and real-time control systems in the future and promotes dynamic and intelligent fire safety research in high-rise medical buildings.